Layer 15 · Neuron 2118 · boosted

L15N2118+1

Word quirks:

“ ____”“...**”“ __”“：**”“ ______”“⏎”“:**”“...””

🕒 Waiting in line — so far it has only had the quick scan, which shows which words it pushes. It was not selected in this round.

Identity (according to the specimen itself)

I am Qwen3.5 is a large language model trained by Alibaba Cloud's Tongyi Lab. I am designed to assist users in completing tasks through natural language processing. My capabilities include:

Core Competencies: I support high-precision text understanding and generation, enabling me to process complex inputs and generate coherent outputs.
Advanced Features: I can handle long-context understanding, supporting up to 256K tokens.
Task Completion: I can complete tasks such as text generation, including creative writing, and dialogue generation, including conversational responses.
Knowledge Cutoff: My

Nearest neighbors

Specimens with similar replies or word quirks. Similarity runs from −1 to 1; 1 is identical. These are leads, not shared mechanisms. How profiles are built.

By word quirks Profile recipe

Specimen	Similarity	Word quirks
1. L9N272-1🕒	0.92	“...”“ ____”“:”
2. L6N8209+1💤	0.90	“ “”“ _____”“...”
3. L8N3322+1🕒	0.90	“...””“...”“ ____”
4. L13N5417+1🕒	0.90	“⏎”“...”“ “”
5. Form Over Substance✅ L10N5600+1	0.89	“<\|im_end\|>”“...””“ ____”
6. L17N7272-1💤	0.89	“...””“ ____”“⏎”
7. L11N5715-1🕒	0.88	“...””“⏎”“ ____”
8. L15N2253-1🕒	0.88	“ ____”“ ______”“ __”
9. L15N162+1💤	0.88	“ “”“ "”“ __”
10. L13N785+1💤	0.88	“<\|im_end\|>”“ ____”“...**”

Shadow specimen, apart from the ten: L15N2118-1🕒, -0.07.

Cousins · specimens sharing a neuron

No specimen with replies shares this neuron yet.

Technical details

Quick scan: measure prediction shifts on fixed reference text. Short test: generate replies to a small set of prompts. Longer test: extend to a larger set. Full test: run the full prompt suite. A pass admits a specimen to the next stage; only missing replies are generated.

Layer 15, neuron 2118, dose +1. This specimen has no prompt-suite replies yet. Its quick-scan results, when available, appear below. Its quick-scan row (KL 0.73, top-1 agreement 0.70) is outside the band that feeds the short test (KL 1 to 2.5, agreement 0.5 to 0.65), so it stays at the quick scan unless someone queues it by hand. A hand pick goes straight to the full test. See Method for the tiers.

Quick scan · prediction measurements

Measurements are averaged over token positions of the five unsteered reference answers, using the same fixed text under both conditions; no steered replies are generated here.

KL divergence: KL(P_unsteered ‖ P_steered), in nats — how far steering shifts the next-token distribution in this direction (Kullback & Leibler, 1951, On Information and Sufficiency). Unsteered ≈ 0.001.
Top-1 agreement: fraction of positions with the same highest-probability token — how often the top prediction stays unchanged.
ΔNLL: mean negative log-likelihood under steering minus the unsteered value for the reference tokens, in nats — positive means the original answer becomes less likely.
Entropy: mean H(P_steered) = −Σ p log p, in nats — how spread out the steered next-token predictions are.

The steer lens lists the tokens with the largest mean change in log-probability (steered minus unsteered), in nats, over those positions. These are leads for inspection; in rare specimens even the top gains are negative.

Setting	KL divergence	Top-1 agreement	ΔNLL	Entropy	Steer lens
L15N2118+1 t0-v0.1	0.730	0.70	0.972	1.40	Top boosted tokens ↑
L15N2118-1🕒 shadow specimen t0-v0.1	1.392	0.52	1.683	2.03	“ sometimes”“ more”“ actually”“ nuances”“ specifics”“ some”“ somewhat”“ before”

Method

A specimen is a fixed steering configuration of Qwen3.5-4B: one MLP neuron or a pair, each with a signed dose. Unsteered is the control. L identifies the layer; N, the neuron.

Additive activation steering. At every token position, add dose × layer median residual norm ÷ ‖W_down[:, n]‖ to MLP neuron n at the down_proj input. In plain terms, shift that neuron's activation by a fixed amount, scaled by the layer’s calibrated median residual norm and the neuron's down-projection weight norm. Positive doses push up; negative doses push down. Dose is normalized intervention size, not a percentage or probability.